OpenAI releases GPT-5 with 'agentic' capabilities as AI companies race toward autonomous systems
OpenAI launched GPT-5, its most capable model to date, featuring extended context, advanced reasoning, and autonomous 'agentic' capabilities that allow the system to plan and execute multi-step tasks without continuous human oversight.
Technology — March 20, 2026
OpenAI released GPT-5, its most capable model to date, in March 2026, introducing what the company described as 'agentic' capabilities that allow the system to autonomously plan and execute sequences of multi-step tasks — from browsing the web and writing code to managing files and interacting with external software services — without requiring continuous human supervision.
CEO Sam Altman said GPT-5 represented 'a qualitative leap beyond what any system has been able to do previously' and described the model as capable of completing work that previously required 'a team of specialists working over days.' The model is being released in a tiered structure, with the most powerful agentic capabilities reserved for enterprise and API customers.
The release reignited a debate about AI safety and autonomous system governance. The Center for AI Safety said GPT-5's agentic design required 'entirely different' risk frameworks from previous language models, noting that systems capable of autonomous multi-step action could cause compounding errors or be exploited in ways that posed novel risks.
Anthropic, Google DeepMind, and xAI all indicated they had comparable releases in preparation. The race to field the first genuinely autonomous commercial AI agent has become the defining competitive dynamic in the industry, with enterprise software customers in finance, law, and consulting identified as the primary addressable market.
Microsoft, which has integrated OpenAI models into its Copilot suite, announced that GPT-5 would be available through Azure as an enterprise agent framework starting in April, with pricing on an agentic task-completion model rather than per-token. Analysts estimated that agentic AI could ultimately replace 10-15% of knowledge worker roles within five years.
The EU's AI Office said it was reviewing GPT-5 against the requirements of the AI Act, noting that some of its agentic capabilities might meet the definition of a 'general-purpose AI model with systemic risk' — a classification that would trigger the Act's most stringent compliance obligations.
OpenAI's valuation was reported at $340 billion following the launch, with the company said to be exploring an IPO for late 2026 or early 2027.
